The paper introduces the Cooperative Classification and Rationalization (C2R) method to improve graph generalization. It addresses challenges in out-of-distribution data by combining classification with rationalization. The approach involves diversifying training distributions and extracting invariant rationales for predictions. Experimental results demonstrate the effectiveness of C2R on both synthetic and real-world datasets.
Graph Neural Networks have shown remarkable achievements in graph classification tasks but struggle with out-of-distribution data. Several approaches have been proposed to address this issue, including diversifying training distributions and extracting invariant rationales for predictions. The Cooperative Classification and Rationalization (C2R) method combines these approaches to enhance graph generalization capabilities. By aligning robust graph representations with rationale subgraph representations, C2R improves model performance on various datasets.
The paper discusses the importance of diverse training distributions and accurate rationale extraction for effective graph generalization. The C2R method integrates these concepts through cooperative learning between classification and rationalization modules. Experimental results validate the effectiveness of C2R in improving model performance on both synthetic and real-world datasets.
To Another Language
from source content
arxiv.org
Principais Insights Extraídos De
by Linan Yue,Qi... às arxiv.org 03-12-2024
https://arxiv.org/pdf/2403.06239.pdfPerguntas Mais Profundas